Retell AI
NewRetell AI is a platform for phone voice agents that answer and place calls, with single-prompt or node-based flows, warm transfers, batch campaigns, simulation testing and pay-as-you-go per-minute pricing.
Retell AI
Explore features, practical uses and pricing below.
Retell AI is a platform for voice agents that answer and place phone calls. Typical jobs are qualifying leads, booking appointments, answering routine questions and passing the caller to a person when needed. Agents are set up in a no-code builder or through the API, and Retell supplies the speech, telephony, testing and monitoring around whichever language model you pick.
Two ways to define a Retell agent
The agent-type guide describes two current options. A single-prompt agent is defined by one prompt, and Retell recommends starting there. A conversation-flow agent splits the call into nodes and transitions, which gives tighter control over steps such as disclosures or identity checks while other replies stay generative. The docs are frank that a flow isn't automatically more reliable; it helps when a long prompt stops being followed or compliance needs fixed wording. Multi-prompt agents are deprecated.
Each agent can run on Retell's own model, Tell-1, or on models from OpenAI, Anthropic and other providers. Voice options include Retell, Cartesia, OpenAI, MiniMax and ElevenLabs voices, each at its own per-minute rate.
Call-handling features worth knowing
- Transfers: cold transfer, warm transfer with an AI summary for the human agent, and a two-way warm transfer where the AI briefs the agent before the caller joins.
- Batch calling: outbound campaigns from a list, kept inside the calling windows you set.
- Telephony: buy numbers from Retell or bring your own carrier over SIP trunking. A Web SDK handles browser calls.
- Review: every call is transcribed, with post-call analysis, live monitoring and an AI QA add-on.
- Integrations: CRM, helpdesk, calendar and file-storage connections, webhooks, an MCP server, and Node.js and Python SDKs.
Testing an after-hours booking line
Take a clinic that wants an agent to book appointments overnight. The team writes a single prompt covering hours, services and booking rules, connects its calendar, and adds a warm transfer to the on-call nurse for anything clinical. Before buying a number, they open simulation testing and write test cases: a caller who wants to reschedule, one who gives the wrong date of birth, one who asks a medical question. Each case has a simulated-caller prompt and success criteria graded pass or fail, and a case passes only if every criterion does. The team runs the cases as a batch and deploys once the batch is clean. After launch they read transcripts and post-call analysis each week and add new test cases from real failures.
Who should look at Retell, and what to check first
Retell fits contact centers, clinics, agencies and software teams that want phone agents without stitching speech, LLM and telephony vendors together themselves. The homepage says Retell is HIPAA and GDPR compliant and SOC 2 Type 1 and Type 2 certified, and it will sign a BAA for healthcare work. Teams that want to choose every provider themselves may prefer Vapi. Check your market on the language support page, and remember that outbound calling carries legal duties around consent and calling hours that the platform's call windows don't settle for you. More platforms like this are in our voice agents category.
How Retell bills by the minute
The pricing page lists pay-as-you-go rates from $0.07 to $0.31 per minute, built from voice infrastructure ($0.055), a TTS voice, the chosen LLM and telephony. New accounts get $10 in free credits and 20 concurrent calls, and extra concurrency costs $8 per call per month. Phone numbers are $2 a month. Add-ons such as PII removal, denoising and knowledge bases carry their own fees, so the real rate depends on the configuration. Enterprise plans add SSO, role-based access control and custom concurrency.
Common Retell AI questions
Can I keep my existing phone number?
Yes. Retell connects to existing telephony over SIP trunking, and the pricing page lists no charge for SIP or custom telephony.
Do I need to write code?
Not for building and testing agents in the dashboard. The API, SDKs and webhooks are there for custom logic.
What happens when the agent can't help?
You set up a transfer to a person, and the builder includes a fallback for when that transfer fails.